12/2/2025, 12:00:00 AM ~ 12/3/2025, 12:00:00 AM (UTC)

Recent Announcements

Announcing Amazon EC2 General purpose M8azn instances (Preview)

Starting today, new general purpose high-frequency high-network Amazon Elastic Compute Cloud (Amazon EC2) M8azn instances are available for preview. These instances are powered by fifth generation AMD EPYC (formerly code named Turin) processors, offering the highest maximum CPU frequency, 5GHz in the cloud. The M8azn instances offer up to 2x compute performance versus previous generation M5zn instances. These instances also deliver 24% higher performance than M8a instances.\n M8azn instances are built on the AWS Nitro System, a collection of hardware and software innovations designed by AWS. The AWS Nitro System enables the delivery of efficient, flexible, and secure cloud services with isolated multitenancy, private networking, and fast local storage. These instances are ideal for applications such as gaming, high-performance computing, high-frequency trading (HFT), CI/CD, and simulation modeling for the automotive, aerospace, energy, and telecommunication industries. To learn more or request access to the M8azn instances preview, visit the Amazon EC2 M8a page.

Announcing Amazon Nova 2 Sonic for real-time conversational AI

Today, Amazon announces the availability of Amazon Nova 2 Sonic, our speech-to-speech model for natural, real-time conversational AI that delivers industry leading quality and price for voice-based conversational AI. It offers best-in-class streaming speech understanding with robustness to background noise and users’ speaking styles, efficient dialog handling, and speech generation with expressive voices that can speak natively in multiple languages (Polyglot voices). It has superior reasoning, instruction following, and tool invocation accuracy over the previous model.\n Nova 2 Sonic builds on the capabilities introduced in the original Nova Sonic model with new features including expanded language support (Portuguese and Hindi), polyglot voices that enable the model to speak different languages with native expressivity using the same voice, and turn-taking controllability to allow developers to set low, medium, or high pause sensitivity. The model also adds cross-modal interaction, allowing users to seamlessly switch between voice and text in the same session, asynchronous tool calling to support multi-step tasks without interrupting conversation flow, and a one-million token context window for sustained interactions.

Developers can integrate Nova Sonic 2 directly into real-time voice systems using Amazon Bedrock’s bidirectional streaming API. Nova Sonic 2 now also seamlessly integrates with Amazon Connect and other leading telephony providers, including Vonage, Twilio, and AudioCodes, as well as open source frameworks such as LiveKit and Pipecat.

Amazon Nova 2 Sonic is available in Amazon Bedrock in the following AWS Regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Stockholm). To learn more, read the AWS News Blog and the Amazon Nova Sonic User Guide. To get started with Nova Sonic 2 in Amazon Bedrock, visit the Amazon Bedrock console.

Announcing the Apache Spark upgrade agent for Amazon EMR

AWS announces the Apache Spark upgrade agent, a new capability that accelerates Apache Spark version upgrades for Amazon EMR on EC2 and EMR Serverless. The agent converts complex upgrade processes that typically take months into projects spanning weeks through automated code analysis and transformation. Organizations invest substantial engineering resources analyzing API changes, resolving conflicts, and validating applications during Spark upgrades. The agent introduces conversational interfaces where engineers express upgrade requirements in natural language, while maintaining full control over code modifications.\n The Apache Spark upgrade agent automatically identifies API changes and behavioral modifications across PySpark and Scala applications. Engineers can initiate upgrades directly from SageMaker Unified Studio, Kiro CLI or IDE of their choice with the help of MCP (Model Context Protocol) compatibility. During the upgrade process, the agent analyzes existing code and suggests specific changes, and engineers can review and approve before implementation. The agent validates functional correctness through data quality validations. The agent currently supports upgrades from Spark 2.4 to 3.5 and maintains data processing accuracy throughout the upgrade process. The Apache Spark upgrade agent is now available in all AWS Regions where SageMaker Unified Studio is available. To start using the agent, visit SageMaker Unified Studio and select IDE Spaces or install the Kiro CLI. For detailed implementation guidance, reference documentation, and migration examples, visit the documentation.

Amazon RDS for SQL Server now supports Developer Edition

Amazon Relational Database Service (Amazon RDS) for SQL Server now offers Microsoft SQL Server 2022 Developer Edition. SQL Server Developer Edition is a free edition of SQL Server that contains all the features of Enterprise Edition and can be used in any non-production environment. This enables customers to build, test, and demonstrate applications using SQL Server while reducing costs and maintaining consistency with their production database configurations.\n Previously, customers that created Amazon RDS for SQL Server instances for development and test environments had to use SQL Server Standard Edition or SQL Server Enterprise Edition, which resulted in additional database licensing costs for non-production usage. Now, customers can lower the cost of their Amazon RDS development and testing instances by using SQL Server Developer Edition. Furthermore, Amazon RDS for SQL Server features such as automated backups, automated software updates, monitoring, and encryption for development and testing purposes will work on Developer Edition. The license for Microsoft SQL Server Developer Edition strictly limits its use to development and testing purposes. It cannot be used in a production environment, or for any commercial purposes that directly serve end-users. For more information, refer to the Amazon RDS for SQL Server User Guide. See Amazon RDS for SQL Server Pricing for pricing details and regional availability.

Amazon S3 Storage Lens adds performance metrics, support for billions of prefixes, and export to S3 Tables

Amazon S3 Storage Lens provides organization-wide visibility into your storage usage and activity to help optimize costs, improve performance, and strengthen data protection. Today, we are adding three new capabilities to S3 Storage Lens that give you deeper insights into your S3 storage usage and application performance: performance metrics that provide insights into how your applications interact with S3 data, analytics for billions of prefixes in your buckets, and metrics export directly to S3 Tables for easier querying and analysis.\n We are adding three specific types of performance metrics. Access pattern metrics identify inefficient requests, including those that are too small and create unnecessary network overhead. Request origin metrics, such as cross-Region request counts, show when applications access data across regions, impacting latency and costs. Object access count metrics reveal when applications frequently read a small subset of objects that could be optimized through caching or moving to high-performance storage. We are expanding the prefix analytics in S3 Storage Lens to enable analyzing billions of prefixes per bucket, whereas previously metrics were limited to the largest prefixes that met minimum size and depth thresholds. This gives you visibility into storage usage and activity across all your prefixes. Finally, we are making it possible to export metrics directly to managed S3 Tables, making them immediately available for querying with AWS analytics services like Amazon QuickSight and enabling you to join this data with other AWS service data for deeper insights. To get started, enable performance metrics or expanded prefixes in your S3 Storage Lens advanced metrics dashboard configuration. These capabilities are available in all AWS Regions, except for AWS China Regions and AWS GovCloud (US) Regions. You can enable metrics export to managed S3 Tables in both free and advanced dashboard configurations in AWS Regions where S3 Tables are available. To learn more, visit the S3 Storage Lens overview page, documentation, S3 pricing page, and read the AWS News Blog.

Amazon FSx for NetApp ONTAP now supports Amazon S3 access

You can now attach Amazon S3 Access Points to your Amazon FSx for NetApp ONTAP file systems so that you can access your file data as if it were in S3. With this new capability, your file data in FSx for NetApp ONTAP is effortlessly accessible for use with the broad range of artificial intelligence, machine learning, and analytics services and applications that work with S3 while your file data continues to reside in your FSx for NetApp ONTAP file system.\n Amazon FSx for NetApp ONTAP is the first and only complete, fully managed NetApp ONTAP file system in the cloud, allowing you to migrate on-premises applications that rely on NetApp ONTAP or other NAS appliances to AWS without having to change how you manage your data. An S3 Access Point is an endpoint that helps control and simplify how different applications or users can access data. Now, with S3 Access Points for FSx for NetApp ONTAP, you can discover new insights, innovate faster, and make even better data-driven decisions with the data you migrate to AWS. For example, you can use your data to augment generative AI applications with Amazon Bedrock, train machine learning models with Amazon SageMaker, run analysis using Amazon Glue or a wide range of AWS Data and Analytics Competency Partner solutions, and run workflows using S3-based cloud-native applications. Get started with this capability by creating and attaching S3 Access Points to new FSx for NetApp ONTAP file systems using the Amazon FSx console, the AWS Command Line Interface (AWS CLI), or the AWS Software Development Kit (AWS SDK). Support for existing FSx for NetApp ONTAP file systems will come in an upcoming weekly maintenance window. This new capability is available in the select AWS Regions. To get started, see the following list of resources:

Amazon FSx for NetApp ONTAP

Amazon S3 Access Points

AWS News Blog

Amazon RDS for SQL Server launches optimize CPU with new generation instances for up to 55% lower price

Amazon RDS for SQL Server launches optimize CPU with support for M7i and R7i instance families, which reduce prices by up to 55% compared to equivalent previous generation instances. Optimize CPU optimizes Simultaneous Multi-threading (SMT) configuration to reduce commercial software charges. Customers can lower cost by upgrading to M7i and R7i instances from similar 6th generation instances. Furthermore, for memory or IO intensive database workloads, customers can get additional cost reduction by fine tuning optimize CPU configuration.\n RDS for SQL Server price for database instance hours consumed is inclusive of Microsoft Windows and Microsoft SQL Server software charges. Optimize CPU disables SMT for instances with 2 or more physical CPU cores. This reduces the number of vCPUs, and the corresponding commercial software charges by 50% while providing the same number of physical CPU cores, and near equivalent performance. The most significant savings are available on 2Xlarge and higher instances, and instances that use Multi-AZ deployment, where RDS optimizes to reduce SQL Server software charges for only a single active node for most usage. For workloads that are memory or IO intensive, customers can fine tune the number of active physical CPU cores for further savings. RDS for SQL Server supports M7i and R7i instances in all AWS Regions. With unbundled instance pricing, database costs are calculated with separate charges for third party licensing fees per vCPU hour, and third party licensing fees are not eligible towards your organization’s discounts with AWS. You can view Microsoft Windows and SQL Server charges associated with your usage on AWS Billing and Cost Management, and in monthly bills. For more details, visit RDS for SQL Server pricing, Amazon RDS User Guide and AWS News Blog.

Amazon EC2 P6e-GB300 UltraServers accelerated by NVIDIA GB300 NVL72 are now generally available

Today, AWS announces the general availability of Amazon Elastic Compute Cloud (Amazon EC2) P6e-GB300 UltraServers. P6e-GB300 UltraServers, accelerated by NVIDIA GB300 NVL72, provide 1.5x GPU memory and 1.5x FP4 compute (without sparsity) compared to P6e-GB200. \n Customers can optimize performance for the most powerful models in production with P6e-GB300 for applications that require higher context and implement emerging inference techniques like reasoning and Agentic AI.

To get started with P6e-GB300 UltraServers, please contact your AWS sales representative.

To learn more about P6e UltraServers and instances, visit Amazon EC2 P6 instances.

Announcing new memory-optimized Amazon EC2 X8aedz Instances

AWS announces Amazon EC2 X8aedz, next generation memory optimized instances, powered by 5th Gen AMD EPYC processors (formerly code named Turin). These instances offer the highest maximum CPU frequency, 5GHz in the cloud. They deliver up to 2x higher compute performance compared to previous generation X2iezn instances.\n X8aedz instances are built using the latest sixth generation AWS Nitro Cards and are ideal for electronic design automation (EDA) workloads such as physical layout and physical verification jobs, and relational databases that benefit from high single-threaded processor performance and a large memory footprint. The combination of 5 GHz processors and local NVMe storage enables faster processing of memory-intensive backend EDA workloads such as floor planning, logic placement, clock tree synthesis (CTS), routing, and power/signal integrity analysis. X8aedz instances feature a 32:1 ratio of memory to vCPU and are available in 8 sizes ranging from 2 to 96 vCPUs with 64 to 3,072 GiB of memory, including two bare metal variants, and up to 8 TB of local NVMe SSD storage. X8aedz instances are now available in US West (Oregon) and Asia Pacific (Tokyo) regions. Customers can purchase X8aedz instances via Savings Plans, On-Demand instances, and Spot instances. To get started, sign in to the AWS Management Console. For more information visit the Amazon EC2 X8aedz instance page or AWS news blog.

Announcing Amazon EC2 Memory optimized X8i instances (Preview)

Amazon Web Services is announcing the preview of Amazon EC2 X8i, next-generation Memory optimized instances. X8i instances are powered by custom Intel Xeon 6 processors delivering the highest performance and fastest memory among comparable Intel processors in the cloud. X8i instances offer 1.5x more memory capacity (up to 6TB) , and up to 3.4x more memory bandwidth compared to previous generation X2i instances.\n X8i instances will be SAP-certified and deliver 46% higher SAPS compared to X2i instances, for mission-critical SAP workloads. X8i instances are a great choice for memory-intensive workloads, including in-memory databases and analytics, large-scale traditional databases, and Electronic Design Automation (EDA). X8i instances offer 35% higher performance than X2i instances with even higher gains for some workloads. To learn more or request access to the X8i instances preview, visit the Amazon EC2 X8i page.

Amazon S3 increases the maximum object size to 50 TB

Amazon S3 increased the maximum object size to 50 TB, a 10x increase from the previous 5 TB limit. This simplifies the processing of large objects such as high-resolution videos, seismic data files, AI training datasets and more. You can store 50 TB objects in all S3 storage classes and use them with all S3 features.\n Optimize upload and download performance for your large objects by using the latest AWS Common Runtime (CRT) and S3 Transfer Manager in the AWS SDK. You can apply S3’s storage management capabilities to these objects. For example, use S3 Lifecycle to automatically archive infrequently accessed objects to S3 Glacier storage classes, or use S3 Replication to copy objects across AWS accounts or Regions. Amazon S3 supports objects up to 50 TB in all AWS Regions. To learn more about working with large objects, visit the S3 User Guide.

Mistral Large 3 and Ministral 3 family now available first on Amazon Bedrock

Customers can now use Mistral Large 3 and the Ministral 3 family of models available first on Amazon Bedrock as well as additional models including Voxtral Mini 1.0, Voxtral Small 1.0, and Magistral Small 1.2 on Amazon Bedrock, a platform for building generative AI applications and agents at production scale.\n Mistral Large 3 is a state-of-the-art, open-weight, general-purpose multimodal model with a granular Mixture-of-Experts architecture featuring 41B active parameters and 675B total parameters, designed for reliability and long-context comprehension. The Ministral 3 family—consisting of 14B, 8B, and 3B models—offers competitive checkpoints across language, vision, and instruct variants, enabling developers to select the right scale for customization and deployment. Amazon Bedrock is the first platform to offer these cutting-edge models, giving customers early access to Mistral AI’s latest innovations. Mistral Large 3 excels at production-grade assistants, retrieval-augmented systems, and complex enterprise workflows with support for a 256K context window and powerful agentic capabilities. The Ministral 3 family complements this with flexible deployment options: Ministral 3 14B delivers advanced multimodal capabilities for local deployment, Ministral 3 8B provides best-in-class text and vision capabilities for edge deployment and single-GPU operation, and Ministral 3 3B offers robust capabilities in a compact package for low-resource environments. Together, these models span the full spectrum from frontier intelligence to efficient edge computing.

These models are now available in Amazon Bedrock. For the full list of available AWS Regions, refer to the documentation.

To get started with these models in Amazon Bedrock, visit the Amazon Bedrock Mistral AI page

Amazon S3 Batch Operations introduces performance improvements

Amazon S3 Batch Operations now completes jobs up to 10x faster at a scale of up to 20 billion objects in a job, helping you accelerate large-scale storage operations.\n With S3 Batch Operations, you can perform operations at scale such as copying objects between staging and production buckets, tagging objects for S3 Lifecycle management, or computing object checksums to verify the content of stored datasets. S3 Batch Operations now pre-processes objects, executes jobs, and generates completion reports up to 10x faster for jobs processing millions of objects with no additional configuration or cost. To get started, create a job in the AWS Management Console and specify operation type as well as filters like bucket, prefix, or creation date. S3 automatically generates the object list, creates an AWS Identity and Access Management (IAM) role with permission policies as needed, then initiates the job. S3 Batch Operations performance improvements are available in all AWS Regions, except for AWS China Regions and AWS GovCloud (US) Regions. For pricing information, please visit the Management & Insights tab of the Amazon S3 pricing page. To learn more about S3 Batch Operations, visit the overview page and documentation.

Announcing Amazon Nova 2 foundation models now available in Amazon Bedrock

Today AWS announces Amazon Nova 2, our next generation of general models that deliver reasoning capabilities with industry-leading price performance. The new models available today in Amazon Bedrock are:\n • Amazon Nova 2 Lite, a fast, cost-effective reasoning model for everyday workloads.

• Amazon Nova 2 Pro (Preview), our most intelligent model for highly complex, multistep tasks.

Amazon Nova 2 Lite and Amazon Nova 2 Pro (Preview) offer significant advancements over our previous generation models. These models support extended thinking with step-by-step reasoning and task decomposition and include three thinking intensity levels—low, medium, and high—giving developers control over the balance of speed, intelligence, and cost. The models also offer built-in tools such as code interpreter and web grounding, support remote MCP tools, and provide a one-million-token context window for richer interactions.

Nova 2 Lite can be used for a broad range of your everyday tasks. It offers the best combination of price, performance, and speed. Early customers are using Nova 2 Lite for customer service chatbots, document processing, and business process automation. Amazon Nova 2 Pro (Preview) can be used for highly complex agentic tasks such as multi-document analysis, video reasoning, and software migrations. Nova 2 Pro is in preview with early access available to all Amazon Nova Forge customers. If interested, reach out to your AWS account team regarding access. Nova 2 Lite can be customized using supervised fine-tuning (SFT) on Amazon Bedrock and Amazon SageMaker, and full fine-tuning is available on Amazon SageMaker.

Amazon Nova 2 Lite and Nova 2 Pro (Preview) is now available in Amazon Bedrock via global cross region inference in multiple locations.

Learn more at the AWS News Blog, Amazon Nova models product page, and Amazon Nova user guide.

Introducing AWS DevOps Agent (preview), frontier agent for operational excellence

We’re excited to launch AWS DevOps Agent in preview, a frontier agent that resolves and proactively prevents incidents, continuously improving reliability and performance of applications in AWS, multicloud, and hybrid environments. AWS DevOps Agent investigates incidents and identifies operational improvements as an experienced DevOps engineer would: by learning your resources and their relationships, working with your observability tools, runbooks, code repositories, and CI/CD pipelines, and correlating telemetry, code, and deployment data across all of them to understand the relationships between your application resources.\n AWS DevOps Agent autonomously triages incidents and guides teams to rapid resolution to reduce Mean Time to Resolution (MTTR). AWS DevOps Agent begins investigating the moment an alert comes in, whether at 2 AM or during peak hours, to quickly restore your application to optimal performance. It analyzes patterns across historical incidents to provide actionable recommendations that strengthen key areas including observability, infrastructure optimization, and deployment pipeline enhancement. AWS DevOps Agent helps access the untapped insights in your operational data and tools without changing your workflows.

AWS DevOps Agent is available at no additional cost during preview in the US East (N. Virginia) Region. To learn more, read the AWS News Blog and see getting started.

Amazon S3 Vectors is now generally available with 40 times the scale of preview

Amazon S3 Vectors, the first cloud object storage with native support to store and query vectors, is now generally available. S3 Vectors delivers purpose-built, cost-optimized vector storage for AI agents, inference, Retrieval Augmented Generation (RAG), and semantic search at billion-vector scale. S3 Vectors is designed to provide the same elasticity, durability, and availability as Amazon S3 and reduces the total costs to upload, store, and query vectors by up to 90%. With general availability, you can store and query up to two billion vectors per index and elastically scale to 10,000 vector indexes per vector bucket. Infrequent queries continue to return results in under one second, with more frequent queries now resulting in latencies around 100 milliseconds or less. Your application can achieve write throughput of 1,000 vectors per second when streaming single-vector updates into your indexes, retrieve up to 100 search results per query, and store up to 50 metadata keys alongside each vector for fine-grained filtering in your queries.\n With S3 Vectors you get a new bucket type—a vector bucket—that is optimized for durable, low-cost vector storage. Within vector buckets, you organize your vector data with vector indexes and get a dedicated set of APIs to store, access, and query vectors without provisioning any infrastructure. By default, S3 Vectors encrypts all vector data in a vector bucket with server-side encryption using S3-managed keys (SSE-S3) or optionally, you can use AWS Key Management Service (SSE-KMS) to set a default customer-managed key to encrypt all new vector indexes in the vector bucket. You can now also set a dedicated customer-managed key per vector index, helping you build scalable multi-tenant applications and meet regulatory and governance requirements. You can also tag vector buckets and indexes for attribute-based access control (ABAC) as well as to track and organize costs using AWS Billing and Cost Management.

S3 Vectors integrates with Amazon Bedrock Knowledge Bases to reduce the cost of using large vector datasets for RAG. When creating a Knowledge Base in Amazon Bedrock or Amazon SageMaker Unified Studio, you can choose an existing Amazon S3 vector index or create a new one using the Quick Create workflow. With Amazon OpenSearch Service, you can optimize costs for hybrid search workloads by configuring OpenSearch to automatically manage vector storage in S3.

S3 Vectors is now generally available in 14 AWS Regions, expanding from 5 Regions in preview. To learn more, visit the product page, S3 pricing page, documentation, and AWS News blog.

Amazon CloudWatch launches unified management and analytics for operational, security, and compliance data

Amazon CloudWatch now provides new data management and analytics capabilities that allow you to unify operational, security, and compliance data across your AWS environment and third-party sources. DevOps teams, security analysts, and compliance officers can now access all their data in a single place, eliminating the need to maintain multiple separate data stores and complex (extract-transform-load) ETL pipelines. CloudWatch now offers greater flexibility in where and how customers gain insights into this data, both natively in CloudWatch or with any Apache Iceberg-compatible tool.\n With the unified data store enhancements, customers can now easily collect and aggregate logs across AWS accounts and regions aligned to geographic boundaries, business units, or persona-specific requirements. With AWS Organization-wide enablement for AWS sources such as AWS CloudTrail, Amazon VPC, and Amazon WAF, and managed collectors for third party sources such as Crowdstrike, Okta, Palo Alto Networks, CloudWatch makes it easy to bring more of your logs together. Customers can use pipelines to transform and enrich their logs to standard formats such as Open Cybersecurity Schema Framework (OCSF) for security analytics, and define facets to accelerate insights on their data. Customers can make their data available in managed Amazon S3 Tables at no additional storage charge, enabling teams to query data in Amazon SageMaker Unified Studio, Amazon Quick Suite, Amazon Athena, Amazon Redshift, or any Apache Iceberg-compatible analytics tool. To get started, visit the Ingestion page in the CloudWatch console and add one or more data sources. To learn more about Amazon CloudWatch unified data store, visit the product page, pricing page, and documentation. For Regional availability, visit the AWS Builder Center.

Announcing Database Savings Plans with up to 35% savings

Today, AWS announces Database Savings Plans, a new flexible pricing model that helps you save up to 35% in exchange for a commitment to a consistent amount of usage (measured in $/hour) over a one-year term with no upfront payment.\n Database Savings Plans automatically apply to eligible serverless and provisioned instance usage regardless of supported engine, instance family, size, deployment option, or AWS Region. For example, with Database Savings Plans, you can change between Aurora db.r7g and db.r8g instances, shift a workload from EU (Ireland) to US (Ohio), modernize from Amazon RDS for Oracle to Amazon Aurora PostgreSQL or from RDS to Amazon DynamoDB and still benefit from discounted pricing offered by Database Savings Plans. Database Savings Plans will be available starting today in all AWS Regions, except China Regions, with support for Amazon Aurora, Amazon RDS, Amazon DynamoDB, Amazon ElastiCache, Amazon DocumentDB (with MongoDB compatibility), Amazon Neptune, Amazon Keyspaces (for Apache Cassandra), Amazon Timestream, and AWS Database Migration Service (DMS). You can get started with Database Savings Plans from the AWS Billing and Cost Management Console or by using the AWS CLI. To realize the largest savings, you can make a commitment to Savings Plans by using purchase recommendations provided in the console. For a more customized analysis, you can use the Savings Plans Purchase Analyzer to estimate potential cost savings for custom purchase scenarios. For more information, visit the Database Savings Plans pricing page and the AWS Savings Plans FAQs.

Amazon Bedrock adds 18 fully managed open weight models, the largest expansion of new models to date

Amazon Bedrock is a platform for building generative AI applications and agents at production scale. Amazon Bedrock provides access to a broad selection of fully managed models from leading AI companies through a unified API, enabling you to evaluate, switch, and adopt new models without rewriting applications or changing infrastructure. Today, Amazon Bedrock is adding 18 fully managed open weight models to its model offering, the largest expansion of new models to date.\n You can now access the following models in Amazon Bedrock:

Google: Gemma 3 4B, Gemma 3 12B, Gemma 3 27B

MiniMax AI: MiniMax M2

Mistral AI: Mistral Large 3, Ministral 3 3B, Ministral 3 8B, Ministral 3 14B, Magistral Small 1.2, Voxtral Mini 1.0, Voxtral Small 1.0

Moonshot AI: Kimi K2 Thinking

NVIDIA: NVIDIA Nemotron Nano 2 9B, NVIDIA Nemotron Nano 2 VL 12B

OpenAI: gpt-oss-safeguard-20b, gpt-oss-safeguard-120b

Qwen: Qwen3-Next-80B-A3B, Qwen3-VL-235B-A22B

For the full list of available AWS Regions, refer to the documentation.

To learn more about all the models that Amazon Bedrock offers, view the Amazon Bedrock model choice page. To get started using these models in Amazon Bedrock, read the launch blog and visit the Amazon Bedrock console.

Amazon Bedrock AgentCore now includes Policy (preview), Evaluations (preview) and more

Today, Amazon Bedrock AgentCore introduces new offerings, including Policy (preview) and Evaluations (preview), to give teams the controls and quality assurance they need to confidently scale agent deployment across their organization, transforming agents from prototypes to solutions in production.\n Policy in AgentCore integrates with AgentCore Gateway to intercept every tool call in real time, ensuring agents stay within defined boundaries without slowing down. Teams can create policies using natural language that automatically convert to Cedar—the AWS open-source policy language—helping development, compliance, and security teams set up, understand, and audit rules without writing custom code. AgentCore Evaluations helps developers test and continuously monitor agent performance based on real-world behavior to improve quality and catch issues before they cause widespread customer impact. Developers can use 13 built-in evaluators for common quality dimensions, such as helpfulness, tools selection, and accuracy, or create custom model-based scoring systems, drastically reducing the effort required to develop evaluation infrastructure. All quality metrics are accessible through a unified dashboard powered by Amazon CloudWatch. We’ve also added new features to AgentCore Memory, AgentCore Runtime, and AgentCore Identity to support more advanced agent capabilities. AgentCore Memory now includes episodic memory, enabling agents to learn and adapt from experiences, building knowledge over time to create more humanlike interactions. AgentCore Runtime supports bidirectional streaming for natural conversations where agents simultaneously listen and respond while handling interruptions and context changes mid-conversation, unlocking powerful voice agent use cases. AgentCore Identity now supports custom claims for enhanced authentication rules across multi-tenant environments while maintaining seamless integration with your chosen identity providers. AgentCore Evaluations is available in preview in four AWS Regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), Europe (Frankfurt). Policy in AgentCore is available in preview in all AWS Regions where AgentCore is available. Learn more about new AgentCore updates through the blog, deep dive using AgentCore resources, and get started with the AgentCore Starter Toolkit. AgentCore offers consumption-based pricing with no upfront costs.

Introducing AWS AI Factories

AWS AI Factories are now available, providing rapidly deployable, high-performance AWS AI infrastructure in your own data centers. By combining the latest AWS Trainium accelerators and NVIDIA GPUs, specialized low-latency networking, high-performance storage, and AWS AI services, AI Factories accelerate your AI buildouts by months or years compared to building independently. Leveraging nearly two decades of AWS cloud leadership expertise, AWS AI Factories eliminate the complexity of procurement, setup, and optimization that typically delays AI initiatives.\n With integrated AWS AI services like Amazon Bedrock and Amazon SageMaker, you gain immediate access to leading foundation models without negotiating separate contracts with individual model providers.  AWS AI Factories operate as dedicated environments built exclusively for you or your designated trusted community, ensuring complete separation and operating independence while integrating with the broader set of AWS services. This approach helps governments and enterprises meet digital sovereignty requirements while benefiting from the unparalleled security, reliability, and capabilities of the AWS Cloud. You provide the data center space and power capacity you’ve already acquired, while AWS deploys and manages the infrastructure. 

AWS AI Factories deliver advanced AI technologies to enterprises across all industries and government organizations seeking secure, isolated environments with strict data residency requirements. These dedicated environments provide access to the same advanced technologies available in public cloud Regions, allowing you to build AI-powered applications as well as train and deploy large language models using your own proprietary data. Rather than spending years building capacity independently, AWS accelerates deployment timelines so you can focus on innovation instead of infrastructure complexity. 

Contact your AWS account team to learn more about deploying AWS AI Factories in your data center and accelerating your AI initiatives with AWS proven expertise in building and maintaining dedicated AI infrastructure at scale.

Amazon S3 Tables now support automatic replication of Apache Iceberg tables

Amazon S3 Tables now support automatic replication of Apache Iceberg tables across AWS Regions and accounts. This new capability replicates your complete table structure, including all snapshots and metadata to reduce query latency and improve data accessibility for global analytics workloads.\n S3 Tables replication automatically creates read-only replica tables in your destination table buckets, backfills them with the latest state of the source table, and continuously monitors for new updates to keep replicas in sync. Replica tables can be configured with independent snapshot retention policies and encryption keys from source tables to meet compliance and data protection requirements. You can query replica tables using Amazon SageMaker Unified Studio or any Iceberg-compatible engine including Amazon Athena, Amazon Redshift, Apache Spark, and DuckDB. S3 Tables replication is now available in all AWS Regions where S3 Tables are supported. For pricing details, visit the Amazon S3 pricing page. To learn more about S3 Tables, visit the product page, documentation, and read the AWS News Blog.

AWS Security Hub is now generally available with near real-time risk analytics

Amazon Web Services (AWS) announces the general availability of AWS Security Hub, a unified cloud security solution that prioritizes your critical security issues and helps you respond at scale, reduce security risks, and improve team productivity. With general availability, Security Hub now includes near real-time risk analytics, advanced trends, unified enablement and management, and streamlined pricing across multiple AWS security services. Security Hub detects critical risks by correlating and enriching security signals from Amazon GuardDuty, Amazon Inspector, and AWS Security Hub CSPM, enabling you to quickly surface and prioritize active risks in your cloud environment.\n Security Hub now delivers near real-time risk analytics and advanced trends, transforming correlated security signals into actionable insights through enhanced visualizations and contextual enrichment. You can enable Security Hub for individual accounts or across your entire AWS Organization with centralized deployment and management. These new capabilities complement existing capabilities, including exposure findings, security-focused resource inventory, attack path visualization, and automated response workflows with ticketing system integration. This centralized management reduces the need for manual correlation across multiple consoles and enables streamlined remediation at scale while helping minimize potential operational disruptions, now with improved cost predictability through streamlined pricing that consolidates charges across multiple AWS security services. The service automatically visualizes potential attack paths by showing how adversaries could chain together threats, vulnerabilities, and misconfigurations to compromise critical resources, providing deeper risk context powered by more comprehensive analytics. For more information about AWS commercial Regions where Security Hub is available, see the AWS Region table. The service integrates with existing AWS security services, providing more comprehensive security posture without additional operational overhead. To learn more about Security Hub and get started, visit the AWS Security Hub console or the AWS Security Hub product page.

Amazon RDS for Oracle and SQL Server now support up to 256 TiB storage with additional storage volumes

Amazon Relational Database Service (Amazon RDS) for Oracle and SQL Server now support up to 256 TiB storage size, a 4x increase in storage size per database instance. Customers can add up to three additional storage volumes in addition to the primary storage volume, each up to 64 TiB storage, to their database instance. Additional storage volumes can be added, scaled up, or removed from the database instance without application downtime, so customers have the flexibility to add and adjust storage volumes over time based on changing workload requirements.\n With additional storage volumes, customers can continue to scale database storage beyond the maximum storage size available in the primary volume. Also, customers can temporarily add volumes when they have a short-term requirement for additional storage, such as for month-end data processing or importing data from local storage, and remove unused volumes when they are no longer required. Furthermore, customers can optimize cost performance by using a combination of high-performance Provisioned IOPS SSD (io2) volumes and General Purpose (gp3) volumes for their database instance. For example, data that requires consistent IOPS performance can be stored on an io2 volume, and infrequently accessed historical data can be stored on a gp3 volume to optimize storage cost. To get started, customers can create additional storage volumes in a new or existing database instance through the AWS Management Console, AWS CLI, or SDKs. For more information, visit the RDS for Oracle User Guide and RDS for SQL Server User Guide. To learn more about how customers can benefit from additional storage volumes, visit the AWS news blog post. Additional storage volumes are available in all commercial AWS Regions and the AWS GovCloud (US) Regions.

Amazon EMR Serverless eliminates local storage provisioning for Apache Spark workloads

Amazon EMR Serverless now offers serverless storage that eliminates local storage provisioning for Apache Spark workloads, reducing data processing costs by up to 20% and preventing job failures from disk capacity constraints. You no longer need to configure local disk type and size for each application. EMR Serverless automatically handles intermediate data operation such as shuffle with no local storage charges. You pay only for compute and memory resources your job consumes.\n EMR Serverless offloads intermediate data operations to a fully managed, auto-scaling serverless storage that encrypts data in transit and at rest with job-level isolation. Serverless storage decouples storage from compute, allowing Spark to release workers immediately when idle rather than keeping workers active to preserve temporary data. It eliminates job failures from insufficient disk capacity and reduces costs by avoiding idle worker charges. This is particularly valuable for jobs using dynamic resource allocation, such as recommendation engines processing millions of customer interactions, where initial stages process large datasets with high parallelism then narrow as data aggregates. This feature is generally available for EMR release 7.12 and later. See Supported AWS Regions for availability. To get started, visit serverless storage for EMR Serverless documentation.

Amazon OpenSearch Service adds GPU-accelerated and auto-optimized vector indexes

You can now build billion-scale vector databases in under an hour on Amazon OpenSearch Service with GPU-acceleration, and auto-optimize vector indexes for optimal trade-offs between search quality, speed and cost.\n Previously, large-scale vector indexes took days to build, and optimizing them required experts to spend weeks of manual tuning. The time, cost and effort weighed down innovation velocity, and customers forwent cost and performance optimizations. You can now run serverless, auto-optimize jobs to generate optimization recommendations. You simply specify search latency and recall requirements, and these jobs will evaluate index configurations (k-NN algorithms, quantization, and engine settings) automatically. Then, you can use vector GPU-acceleration to build an optimized index up to 10X faster at a quarter of the indexing cost. Serverless GPUs dynamically activate and accelerate your domain or collection, so you’re only billed when you benefit from speed boosts—all done without you managing GPU instances. These capabilities help you scale AI applications including semantic search, recommendation engines, and agentic systems more efficiently. By simplifying and accelerating the time to build large-scale, optimized vector databases, your team will be empowered to innovate faster. Vector GPU-acceleration is available for vector collections and OpenSearch 3.1+ domains in US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), Europe (Ireland), and Asia Pacific (Tokyo) Regions. Vector auto-optimize is available for vector collections and OpenSearch 2.17+ domains in US East (Ohio), US East (N. Virginia), US West (Oregon), Asia Pacific (Mumbai), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), Europe (Frankfurt) and Europe (Ireland) Regions. Learn more.

Build agents to automate production UI workflows with Amazon Nova Act (GA)

We are excited to announce the general availability of Amazon Nova Act, a new AWS service for developers to build and manage fleets of highly reliable agents for automating production UI workflows. Nova Act is powered by a custom Nova 2 Lite model and provides high reliability with unmatched cost efficiency, fastest time-to-value, and ease of implementation at scale.\n Nova Act can reliably complete repetitive UI workflows in the browser, execute APIs or tools (e.g. write to PDF), and escalate to a human supervisor when appropriate. Developers that need to automate repetitive processes across the enterprise can define workflows combining the flexibility of natural language with more deterministic Python code. Technical teams using Nova Act can start prototyping quickly on the online playground at nova.amazon.com/act, refine and debug their scripts using the Nova Act IDE extension, and deploy to AWS in just a few steps. Nova Act is available today in AWS Region US East (N. Virginia). Learn more about Nova Act.

AWS previews EC2 C8ine instances

AWS launches the preview of Amazon EC2 C8ine instances, powered by custom sixth-generation Intel Xeon Scalable processors (Granite Rapids) and the latest AWS Nitro v6 card. These instances are designed specifically for dataplane packet processing workloads.\n Amazon EC2 C8ine instance configurations can deliver up to 2.5 times higher packet performance per vCPU versus prior generation C6in instances. They can offer up to 2x higher network bandwidth through internet gateways and up to 3x more Elastic Network Interface (ENI) compared to existing C6in network optimized instances. They are ideal for packet processing workloads requiring high performance at small packet sizes. These workloads include security virtual appliances, firewalls, load balancers, DDoS protection systems, and Telco 5G UPF applications. These instances are available for preview upon request through your AWS account team. Connect with your account representatives to signup.

Amazon GuardDuty Extended Threat Detection now supports Amazon EC2 and Amazon ECS

AWS announces further enhancements to Amazon GuardDuty Extended Threat Detection with new capabilities to detect multistage attacks targeting Amazon Elastic Compute Cloud (Amazon EC2) instances and Amazon Elastic Container Service (Amazon ECS) clusters running on AWS Fargate or Amazon EC2. GuardDuty Extended Threat Detection uses artificial intelligence and machine learning algorithms trained at AWS scale to automatically correlate security signals and detect critical threats. It analyzes multiple security signals across network activity, process runtime behavior, malware execution, and AWS API activity over extended periods to detect sophisticated attack patterns that might otherwise go unnoticed.\n With this launch, GuardDuty introduces two new critical-severity findings: AttackSequence:EC2/CompromisedInstanceGroup and AttackSequence:ECS/CompromisedCluster. These findings provide attack sequence information, allowing you to spend less time on initial analysis and more time responding to critical threats, minimizing business impact. For example, GuardDuty can identify suspicious processes followed by persistence attempts, crypto-mining activities, and reverse shell creation, representing these related events as a single, critical-severity finding. Each finding includes a detailed summary, events timeline, mapping to MITRE ATT&CK® tactics and techniques, and remediation recommendations. While GuardDuty Extended Threat Detection is automatically enabled for GuardDuty customers at no additional cost, its detection comprehensiveness depends on your enabled GuardDuty protection plans. To improve attack sequence coverage and threat analysis of Amazon EC2 instances, enable Runtime Monitoring for EC2. To enable detection of compromised ECS clusters, enable Runtime Monitoring for Fargate or EC2 depending on your infrastructure type. To get started, enable GuardDuty protection plans via the Console or API. New GuardDuty customers can start with a 30-day free trial, and existing customers who haven’t used Runtime Monitoring can also try it free for 30 days. For additional information, visit the blog post and Amazon Guard Duty product page.

Announcing Amazon EC2 Trn3 UltraServers for faster, lower-cost generative AI training

AWS announces the general availability of Amazon Elastic Compute Cloud (Amazon EC2) Trn3 UltraServers powered by our fourth–generation AI chip Trainium3, our first 3nm AWS AI chip purpose-built to deliver the best token economics for next-generation agentic, reasoning, and video generation applications.\n Each AWS Trainium3 chip provides 2.52 petaflops (PFLOPs) of FP8 compute, increases the memory capacity by 1.5x and bandwidth by 1.7x over Trainium2 to 144 GB of HBM3e memory, and 4.9 TB/s of memory bandwidth. Trainium3 is designed for both dense and expert-parallel workloads with advanced data types (MXFP8 and MXFP4) and improved memory-to-compute balance for real-time, multimodal, and reasoning tasks. Trn3 UltraServers can scale up to 144 Trainium3 chips (362 FP8 PFLOPs total) and are available in EC2 UltraClusters 3.0 to scale to hundreds of thousands of chips. A fully configured Trn3 UltraServer delivers up to 20.7 TB of HBM3e and 706 TB/s of aggregate memory bandwidth. The next-generation Trn3 UltraServer, feature the NeuronSwitch-v1, an all-to-all fabric that doubles interchip interconnect bandwidth over Trn2 UltraServer. Trn3 delivers up to 4.4x higher performance, 3.9x higher memory bandwidth and 4x better performance/watt compared to our Trn2 UltraServers, providing the best price-performance for training and serving frontier-scale models, including reinforcement learning, Mixture-of-Experts (MoE), reasoning, and long-context architectures. On Amazon Bedrock, Trainium3 is our fastest accelerator, delivering up to 3× faster performance than Trainium2 with over 5× higher output tokens per megawatt at similar latency per user. New Trn3 UltraServers are built for AI researchers and powered by the AWS Neuron SDK, to unlock breakthrough performance. With native PyTorch integration, developers can train and deploy without changing a single line of model code. For AI performance engineers, we’ve enabled deeper access to Trainium3 so they can fine-tune performance, customize kernels, and push models even further. Because innovation thrives on openness, we are committed to engaging with our developers through open-source tools and resources.

AWS Lambda announces durable functions for multi-step applications and AI workflows

AWS Lambda announces durable functions, enabling developers to build reliable multi-step applications and AI workflows within the Lambda developer experience. Durable functions automatically checkpoint progress, suspend execution for up to one year during long-running tasks, and recover from failures - all without requiring you to manage additional infrastructure or write custom state management and error handling code.\n Customers use Lambda for the simplicity of its event-driven programming model and built-in integrations. While traditional Lambda functions excel at handling single, short-lived tasks, developers building complex multi-step applications, such as order processing, user onboarding, and AI-assisted workflows, previously needed to implement custom state management logic or integrate with external orchestration services. Lambda durable functions address this opportunity by extending the Lambda programming model with new operations like “steps” and “waits” that let you checkpoint progress and pause execution without incurring compute charges. The service handles state management, error recovery, and efficient pausing and resuming of long-running tasks, allowing you to focus on your core business logic. Lambda durable functions are generally available in US East (Ohio) with support for Python (versions 3.13 and 3.14) and Node.js (versions 22 and 24) runtimes. For the latest region availability, visit the AWS Capabilities by Region page. You can activate durable functions for new Python or Node.js based Lambda functions using the AWS Lambda API, AWS Management Console, AWS Command Line Interface (AWS CLI), AWS Cloud Formation, AWS Serverless Application Model (AWS SAM), AWS SDK, and AWS Cloud Development Kit (AWS CDK). For more information on durable functions, visit the AWS Lambda Developer Guide and launch blog post. To learn about pricing, visit AWS Lambda pricing.

AWS Support transformation: AI-powered operations with the human expertise you trust

AWS Support announces a transformation of its Support portfolio, simplified into three intelligent, experience-driven plans: Business Support+, Enterprise Support, and Unified Operations. Each plan combines the speed and precision of AI with the expertise of AWS engineers. Each higher plan builds on the previous one, adding faster response times, proactive guidance, and smarter operations. The result: reduced engineering burden, stronger reliability and resiliency, and streamlined cloud operations.\n Business Support+ delivers 24/7 AI-powered assistance that understands your context, with direct engagement to AWS experts for critical issues within 30 minutes—twice as fast as current plans. Enterprise Support expands on this with designated Technical Account Managers (TAMs) who blend generative AI insights with human judgment to provide strategic operational guidance across resiliency, cost, and efficiency. It also includes AWS Security Incident Response at no additional cost, which customers can activate to automate security alert investigation and triage. Unified Operations, the top plan, is designed for mission-critical workloads—offering a global team of designated experts who deliver architecture reviews, guided testing, proactive optimization, and five-minute context-specific response times for critical incidents. Customers using AWS DevOps Agent (preview) can engage with AWS Support with one-click from an investigation when needed, giving AWS experts immediate context for faster resolution. AWS DevOps Agent is a frontier agent that resolves and proactively prevents incidents, continuously improving reliability and performance of applications in AWS, multicloud, and hybrid environments. Business Support+, Enterprise Support, and Unified Operations are available in all commercial AWS Regions. Existing customers can continue with their current plans or explore the new offerings for enhanced performance and efficiency. To see how AWS blends AI intelligence and human expertise to transform your cloud operations, visit the AWS Support product page.

AWS Security Agent (Preview): AI agent for proactive app security

Today, AWS announces the preview of AWS Security Agent, an AI-powered agent that proactively secures your applications throughout the development lifecycle. AWS Security Agent conducts automated security reviews tailored to your organizational requirements and delivers context-aware penetration testing. By continuously validating security from design to deployment, it helps prevent vulnerabilities early in development across all your environments.\n Security teams define organizational security requirements once in the AWS Security Agent console, such as approved encryption libraries, authentication frameworks, and logging standards. AWS Security Agent then automatically validates these requirements throughout development by evaluating architectural documents and code against your defined standards, providing specific guidance when violations are detected. For deployment validation, security teams define their penetration testing scope and AWS Security Agent develops application context, executes sophisticated attack chains, and discovers and validates vulnerabilities. This delivers consistent security policy enforcement across all teams, scales security reviews to match development velocity, and transforms penetration testing from a periodic bottleneck into an on-demand capability that dramatically reduces risk exposure.

AWS Security Agent (Preview) is currently available in the US East (N. Virginia) Region. All of your data remains safe and private. Your queries and data are never used to train models. AWS Security Agent logs API activity to AWS CloudTrail for auditing and compliance.

To learn more about AWS Security Agent, visit the product page and read the launch announcement. For technical details and to get started, see the AWS Security Agent documentation.

Announcing New Compute-Optimized Amazon EC2 C8a Instances

AWS announces the general availability of new compute-optimized Amazon EC2 C8a instances. C8a instances are powered by 5th Gen AMD EPYC processors (formerly code named Turin) with a maximum frequency of 4.5 GHz, delivering up to 30% higher performance and up to 19% better price-performance compared to C7a instances.\n C8a instances deliver 33% more memory bandwidth compared to C7a instances, making these instances ideal for latency sensitive workloads. Compared to Amazon EC2 C7a instances, they are up to 57% faster for GroovyJVM allowing better response times for Java-based applications. C8a instances offer 12 sizes including 2 bare metal sizes. This range of instance sizes allows customers to precisely match their workload requirements. C8a instances are built on AWS Nitro System and are ideal for high performance, compute-intensive workloads such as batch processing, distributed analytics, high performance computing (HPC), ad serving, highly-scalable multiplayer gaming, and video encoding. C8a instances are available in the following AWS Regions: US East (N. Virginia), US East (Ohio), and US West (Oregon) regions. To get started, sign in to the AWS Management Console. Customers can purchase these instances via Savings Plans, On-Demand instances, and Spot instances. For more information visit the Amazon EC2 C8a instance page.

Announcing Amazon EC2 M4 Max Mac instances (Preview)

Amazon Web Services announces preview of Amazon EC2 M4 Max Mac instances, powered by the latest Mac Studio hardware. Amazon EC2 M4 Max Mac instances are the next-generation EC2 Mac instances, that enable Apple developers to migrate their most demanding build and test workloads onto AWS. These instances are ideal for building and testing applications for Apple platforms such as iOS, macOS, iPadOS, tvOS, watchOS, visionOS, and Safari.\n M4 Max Mac instances are powered by the AWS Nitro System, providing up to 10 Gbps network bandwidth and 8 Gbps of Amazon Elastic Block Store (Amazon EBS) storage bandwidth. These instances are built on Apple M4 Max Mac Studio computers featuring a 16-core CPU, 40-core GPU, 16-core Neural Engine, and 128GB of unified memory. Compared to EC2 M4 Pro Mac instances, M4 Max instances offer twice the GPU cores and more than 2.5x the unified memory, offering customers more choice to match instance capabilities to their specific workload requirements and further expanding the selection of Apple silicon Mac hardware on AWS.

To learn more or request access to the Amazon EC2 M4 Max Mac instances preview, visit the Amazon EC2 Mac page.

Amazon Nova Forge: Build your own Frontier Models using Nova

We are excited to announce the general availability of Nova Forge, a new service to build your own frontier models using Nova.\n With Nova Forge, you can start your model development on SageMaker AI from early Nova checkpoints across pre-training, mid-training, or post-training phases. You can blend proprietary data with Amazon Nova-curated data to train the model. You can also take advantage of model development features available exclusively on Nova Forge, including the ability to execute Reinforcement Fine Tuning (RFT) with reward functions in your environment and to implement custom safety guardrails using the built-in responsible AI toolkit. Nova Forge allows you to build models that deeply understand your organization’s proprietary knowledge and reflects your expertise, while preserving general capabilities like reasoning and minimizing risks like catastrophic forgetting. In addition, Nova Forge customers get early access to new Nova models, including Nova 2 Pro and Nova 2 Omni. Nova Forge is available today in US East (N. Virginia) AWS Region and will be available in additional regions in the coming months. Learn more about Nova Forge on the AWS News Blog, the Amazon Nova Forge product page, or the Amazon Nova Forge user guide. You can get started with Nova Forge today from the Amazon Nova Forge console.

Introducing Amazon Nova 2 Omni in Preview

We are excited to announce Amazon Nova 2 Omni, an all-in-one model for multimodal reasoning and image generation. It is the industry’s first reasoning model that supports text, images, video, and speech inputs while generating both text and image outputs. It enables multimodal understanding, image generation and editing using natural language, and speech transcription.\n Unlike traditional approaches that often force organizations to stitch together various specialized models, each supporting different input and output types, Nova 2 Omni eliminates the complexity of managing multiple AI models. This helps to accelerate application development while reducing complexity and costs, enabling developers to tackle diverse tasks from marketing content creation and customer support call transcription to video analysis and documentation with visual aids. The model supports a 1M token context window, 200+ languages for text processing and 10 languages for speech input. It can generate and edits high-quality images using natural language, enabling character consistency, text rendering within image as well as object and background modification. Nova 2 Omni delivers superior speech understanding with native reasoning to transcribe, translate and summarize multi-speaker conversations. And with flexible reasoning controls for depth and budget, developers can ensure optimal performance, accuracy, and cost management across different use cases. Nova 2 Omni is in preview with early access available to all Nova Forge customers. Please reach out to your AWS account team for access. To learn more about Amazon Nova 2 Omni read the user guide.

Amazon API Gateway adds MCP proxy support

Amazon API Gateway now supports Model Context Protocol (MCP) proxy, allowing you to transform your existing REST APIs into MCP-compatible endpoints. This new capability enables organizations to make their APIs accessible to AI agents and MCP clients. Through integration with Amazon Bedrock AgentCore’s Gateway service, you can securely convert your REST APIs into agent-compatible tools while enabling intelligent tool discovery through semantic search.\n The MCP proxy capability, alongside Bedrock AgentCore Gateway services, delivers three key benefits. First, it enables REST APIs to communicate with AI agents and MCP clients through protocol translation, eliminating the need for application modifications or managing additional infrastructure. Second, it provides comprehensive security through dual authentication - verifying agent identities for inbound requests while managing secure connections to REST APIs for outbound calls. Finally, it enables AI agents to search and select the most relevant REST APIs that best match the prompt context. To learn about pricing for this feature, please see the Amazon Bedrock AgentCore pricing page. Amazon API Gateway MCP proxy capability is available in the nine AWS Regions that Amazon Bedrock AgentCore is available in: Asia Pacific (Mumbai), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), Europe (Dublin), Europe (Frankfurt), US East (N. Virginia), US East (Ohio), and US West (Oregon). To get started, visit Amazon API Gateway documentation.

Amazon SageMaker Catalog now exports asset metadata as queryable dataset

Amazon SageMaker Catalog now exports asset metadata as an Apache Iceberg table through Amazon S3 Tables. This allows data teams to query catalog inventory and answer questions such as, “How many assets were registered last month?”, “Which assets are classified as confidential?”, or “Which assets lack business descriptions?” using standard SQL without building custom ETL infrastructure for reporting.\n This capability automatically converts catalog asset metadata into a queryable table accessible from Amazon Athena, SageMaker Unified Studio notebooks, AI agents, and other analytics and BI tools. The exported table includes technical metadata (such as resource_id, resource_type), business metadata (such as asset_name, business_description), ownership details, and timestamps. Data is partitioned by snapshot_date for time travel queries and automatically appears in SageMaker Unified Studio under the aws-sagemaker-catalog bucket.

This capability is available in all AWS Regions where SageMaker Catalog is supported at no additional charge. You pay only for underlying services including S3 Tables storage and Amazon Athena queries. You can control storage costs by setting retention policies on the exported tables to automatically remove records older than your specified period. To get started, activate dataset export using the AWS CLI, then access the asset table through S3 Tables or SageMaker Unified Studio’s Data tab within 24 hours. Query using Amazon Athena, Studio notebooks, or connect external BI tools through the S3 Tables Iceberg REST Catalog endpoint. For instructions, see the Amazon SageMaker user guide.

Amazon CloudWatch GenAI observability now supports Amazon AgentCore Evaluations

Amazon CloudWatch now enables automated quality assessment of AI agents through AgentCore Evaluations. This new capability helps developers continuously monitor and improve agent performance based on real-world interactions, allowing teams to identify and address quality issues before they impact customers.\n AgentCore Evaluations comes with 13 pre-built evaluators covering essential quality dimensions like helpfulness, tool selection, and response accuracy, while also supporting custom model-based scoring systems. You can access unified quality metrics and agent telemetry in CloudWatch dashboards, with end-to-end tracing capabilities to correlate evaluation metrics with prompts and logs. The feature integrates seamlessly with CloudWatch’s existing capabilities including Application Signals, Alarms, Sensitive Data Protection, and Logs Insights. This capability eliminates the need for teams to build and maintain custom evaluation infrastructure, accelerating the deployment of high-quality AI agents. Developers can monitor their entire agent fleet through the AgentCore section in the CloudWatch GenAI observability console.

AgentCore Evaluations is now available in US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney). To get started, visit the documentation and pricing details. Standard CloudWatch pricing applies for underlying telemetry data.

Announcing new memory optimized Amazon EC2 X8aedz instances

AWS announces Amazon EC2 X8aedz, next generation memory optimized instances, powered by 5th Gen AMD EPYC processors (formerly code named Turin). These instances offer the highest maximum CPU frequency, 5GHz in the cloud. They deliver up to 2x higher compute performance compared to previous generation X2iezn instances.\n X8aedz instances are built using the latest sixth generation AWS Nitro Cards and are ideal for electronic design automation (EDA) workloads such as physical layout and physical verification jobs, and relational databases that benefit from high single-threaded processor performance and a large memory footprint. The combination of 5 GHz processors and local NVMe storage enables faster processing of memory-intensive backend EDA workloads such as floor planning, logic placement, clock tree synthesis (CTS), routing, and power/signal integrity analysis. X8aedz instances feature a 32:1 ratio of memory to vCPU and are available in 8 sizes ranging from 2 to 96 vCPUs with 64 to 3,072 GiB of memory, including two bare metal variants, and up to 8 TB of local NVMe SSD storage. X8aedz instances are now available in US West (Oregon) and Asia Pacific (Tokyo) regions. Customers can purchase X8aedz instances via Savings Plans, On-Demand instances, and Spot instances. To get started, sign in to the AWS Management Console. For more information visit the Amazon EC2 X8aedz instance page or AWS news blog.

Amazon Bedrock AgentCore Runtime now supports bi-directional streaming

Amazon Bedrock AgentCore Runtime now supports bi-directional streaming, enabling real-time conversations where agents listen and respond simultaneously while handling interruptions and context changes mid-conversation. This feature eliminates conversational friction by enabling continuous, two-way communication where context is preserved throughout the interaction.\n Traditional agents require users to wait for them to finish responding before providing clarification or corrections, creating stop-start interactions that break conversational flow and feel unnatural, especially in voice applications. Bi-directional streaming addresses this limitation by enabling continuous context handling, helping power voice agents that deliver natural conversational experiences where users can interrupt, clarify, or change direction mid-conversation, while also enhancing text-based interactions through improved responsiveness. Built into AgentCore Runtime, this feature eliminates months of engineering effort required to build real-time streaming capabilities, so developers can focus on building innovative agent experiences rather than managing complex streaming infrastructure. This feature is available in all nine AWS Regions where Amazon Bedrock AgentCore Runtime is available: US East (N. Virginia), US East (Ohio), US West (Oregon), Asia Pacific (Mumbai), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), Europe (Frankfurt), and Europe (Ireland). To learn more about AgentCore Runtime bi-directional streaming, read the blog, visit the AgentCore documentation and get started with the AgentCore Starter Toolkit. With AgentCore Runtime’s consumption-based pricing, you only pay for active resources consumed during agent execution, with no charges for idle time or upfront costs.

Amazon SageMaker AI announces serverless MLflow capability for faster AI development

Amazon SageMaker AI now offers a serverless MLflow capability that dynamically scales to support AI model development tasks. With MLflow, AI developers can begin tracking, comparing, and evaluating experiments without waiting for infrastructure setup.\n As customers across industries accelerate AI development, they require capabilities to track experiments, observe behavior, and evaluate the performance of AI models, applications and agents. However, managing MLflow infrastructure requires administrators to continuously maintain and scale tracking servers, make complex capacity planning decisions, and deploy separate instances for data isolation. This infrastructure burden diverts resources away from core AI development and creates bottlenecks that impact team productivity and cost effectiveness. With this update, MLflow now scales dynamically to deliver fast performance for demanding and unpredictable model development tasks, then scales down during idle time. Administrators can also enhance productivity by setting up cross-account access via Resource Access Manager (RAM) to simplify collaboration across organizational boundaries. The serverless MLflow capability on Amazon SageMaker AI is offered at no additional charge and works natively with familiar Amazon SageMaker AI model development capabilities like SageMaker AI JumpStart, SageMaker Model Registry and SageMaker Pipelines. Customers can access the latest version of MLflow on Amazon SageMaker AI with automatic version updates. Amazon SageMaker AI with MLflow is now available in select AWS Regions. To learn more, see the Amazon SageMaker AI user guide and the AWS News Blog.

Amazon S3 Tables now offer the Intelligent-Tiering storage class

Amazon S3 Tables now offer the Intelligent-Tiering storage class, which optimizes costs based on access patterns, without performance impact or operational overhead. Intelligent-Tiering automatically transitions data in tables across three low-latency access tiers as access patterns change, reducing storage costs by up to 80%. Additionally, S3 Tables automated maintenance operations such as compaction, snapshot expiration, and unreferenced file removal never tier up your data. This helps you to keep your tables optimized while saving on storage costs.\n With the Intelligent-Tiering storage class, data in tables not accessed for 30 consecutive days automatically transitions to the Infrequent Access tier (40% lower cost than the Frequent Access tier). After 90 days without access, that data transitions to the Archive Instant Access tier (68% lower cost than the Infrequent Access tier). You can now select Intelligent-Tiering as the storage class when you create a table or set it as the default for all new tables in a table bucket. The Intelligent-Tiering storage class is available in all AWS Regions where S3 Tables are available. For pricing details, visit the Amazon S3 pricing page. To learn more about S3 Tables, visit the product page, documentation, and read the AWS News Blog.

AWS Blogs

AWS Japan Blog (Japanese)

AWS News Blog

Open Source Project

AWS CLI

Amplify for iOS